Data processing systems are frequently comprised of a plurality of client platforms, such as personal workstations or personal computers, connected through networks to one or more server platforms which provide data related services to the application programs executing on the client platforms. The data related services may include data storage and retrieval, data protection, and electronic mail services and such services may be provided to the users from both local servers and from remote servers which are networked to a user's local server.
A number of problems arise from such system configurations, however, one being that the client and server platforms are frequently based upon different operating system. For example, the client platforms may use Microsoft Windows and application programs designed to use Microsoft Windows while the server platforms may be based upon the UNIX operating system. As such, the connection and communications between the client platforms and the server platforms must be of a nature to be compatible with both types of operating systems and associated application and services programs.
Other problems arise from the inherent limitations of the connection and communications facilities associated with the client applications and, as a separable problem, with the inherent limitations of the server programs, such as the data storage and retrieval programs executing in the server platforms. These problems severely limit the capabilities of the client platforms and server platforms to communicate and to execute data storage and retrieval operations.
Referring first to the client platforms, client platforms are frequently limited in the number of network connections that they can support while there is traditionally one network connection for each client application, even if the connections are to the same server task. This in turn rapidly uses up the available client connections that can be supported by the client platform and results and a significantly slower startup time for each application when it attempts to connect to a server as a given client application may have to wait until a connection is established.
In addition, certain applications, such as those using Microsoft Windows, are pseudo multitasking rather than true multitasking, so that only the application currently having the operating system context can send and receive messages, and are non-preemptive, so that the current application will complete all message operations before passing the context to another application, so that only one application may make use of the connections at a time. Still further, such applications may be synchronous in that they will send a message or a request for an operation and then will wait until a response is received before executing a next operation. Therefore, not only are the available connections rapidly used up, but a given application may significantly delay other applications access to the available connections by forcing the other applications to wait until the application having a connection completes all of its operations.
One solution of the prior art to this problem has been to provide a connection sharing architecture, usually based upon a semaphore mechanism used in common by the applications to indicate when a connection is free for use by another application. This approach, however, not only does not solve all of the problems of the prior art as described above, but places a further burden on the application programs in that each application program must know of the connection sharing mechanism and must operate with the mechanism. This, for example, requires each application to deal with semaphoring and to hold, or queue, requests until a connection is free.
Referring now to the server platforms, server platforms usually provide a server task which operates alone to service requests one at a time. This in turn requires that the server task queue or otherwise hold pending requests until the server task has completely finished with each prior request.
One solution of the prior art has been to start a new server task or process for each new connection to the server wherein each process handles requests only from it own connection. This approach, however, substantially increases connection startup times because a new server process must be started for each new connection. In addition, this approach uses server resources inefficiently because a server process is idle until a request appears on its connection and, because an individual connection in a client/server model typically does not have frequent activity, the associated server process will be idle most of the time.
Another solution of the prior art has been for the server to include a dispatcher task which performs preliminary operations upon each incoming request and then passes the parameters of the request through an interprocess communication facility to a worker task to process. This approach is limited, however, in that the number of operations that the dispatcher must perform for each request limits the number of requests that the dispatcher can process in a given time. That is, when the rate at which requests are submitted to the server exceeds the rate at which the dispatcher can process the requests, the delay time in responding to a given request will increase to the point where the time of the server is unacceptable. As such, the rate at which requests are submitted to the dispatcher must be limited, for example, by limiting the number of connections supported by the dispatcher or by limiting the rate at which requests may be submitted through the connections. In addition, the dispatcher is not available to detect new requests while processing a current request, thereby requiring a queue mechanism to hold new requests for the dispatcher. These problems are compounded in that the request parameters frequently include addresses, thereby requiring the dispatcher task to perform address resolution operations and further slowing the processing of requests by the server.
Finally, yet other problems in systems of the prior art arise from providing system security, usually by checking the access authorizations of user to various system resources, such as databases and electronic mail services. For example, one well known and often used authorization mechanism of the prior art involves an authentication server and a directory server wherein the directory server stores the authorization rights of the clients to various system resources and a set of individual passwords for the clients and for the system resources. The client makes a request to the authentication server for an identification packet which identifies the client and the authentication server provides a corresponding identification packet containing an identification of the client and this identification packet is encrypted using the password of the server as the encryption key. The client then sends the identification packet to the server, which decodes the identification packet with its password to obtain the identification of the client and uses this client identification to access the directory server to obtain the authorization rights of the client. This approach, however, places substantial burdens on both the directory server and the server which has been accessed, due to the number of directory access operations. In addition, this approach presents serious potential security problems in that all servers must have access to the directory server and must therefore be trusted, so that a false server could penetrate system security.